Bird image classification based on joint, the inception V3 model, and transfer learning

Bird classification is essential in zoology, computer vision, and artificial intelligence. This paper combines the Inception V3 model with transfer learning to create an effective method for classifying bird images. In addition, this paper observes how the model is affected by the number of epochs. Some top convolutional layers are fine-tuned to perform better after training the fully connected layers. Meanwhile, training the model with more epochs to observe the influence of epochs on model loss and accuracy. This research is conducted on the "BIRDS 525 SPECIES" dataset, and the results indicate that the new model achieves good classification results with relatively few epochs. Fine-tuning the model further enhances its generalization ability, and it is observed that due to maintaining most weights unchanged in transfer learning, more epochs cause the model to converge without a decrease in accuracy, effectively addressing overfitting issues. The combination of transfer learning and convolutional neural networks allows the model to perform well on small datasets at a lower cost while improving overfitting issues, making it a promising approach for future applications on various platforms and small-scale devices.